EUROPEAN WINTER STORMS: DYNAMICAL
ASPECTS AND WIND GUST ESTIMATION BASED ON
RESULTS OF REGIONAL CLIMATE MODEL
SIMULATIONS
Inaugural- Dissertation
zur
Erlangung des Doktorgrades
der Mathematisch-Naturwissenschaftlichen Fakultät
der Universität zu Köln
vorgelegt von
Patrick Ludwig
aus Mönchengladbach
Köln, 2015
Berichterstatter: Prof. Dr. Michael Kerschgens Prof. Dr. Andreas Fink Tag der mündlichen Prüfung: 02.06.2014
Abstract .
Abstract
Extratropical cyclones in the North Atlantic – European sector are among the most perilous and damaging natural hazards affecting Europe. While most of the severe extratropical cyclones pass by Europe in northeastern direction, a small number of strong storms hit Europe each year. Their destructive power is mainly related to strong wind gusts, sustained high wind speeds or huge amounts of precipitation. Especially the relation between wind gusts and losses is a current topic of research. The focus of this thesis is to analyse severe extratropical cyclones affecting Europe during the winter half year (winter storms). The investigation of dynamical aspects and mesoscale processes associated with these hazardous extratropical cyclones is based on results from partly high-resolution mesoscale modelling approaches with the regional climate model COSMO-CLM. In the first part of this study, the ability of the COSMO-CLM to simulate severe winter storm events realistic is verified. With this aim, a total of 158 historical winter storms events between 1972 and 2008 are simulated. A new physically based wind gust estimation method, extended by a probabilistic approach, has been implemented to the COSMO-CLM to provide realistic area- wide wind gust distributions during the storm passage. In the second part, two recent severe winter storms (Kyrill in January 2007, Xynthia in February 2010) that caused widespread damage and even fatalities are investigated in more detail. Particularly, the dynamical aspects and mesoscale processes affecting their development are considered. In general, the results approve the ability of realistic simulations of severe winter storm events by the COSMO-CLM. Further, the novel introduced wind gust estimation method provides comparable results to existing wind gust estimation methods. The probabilistic extension permits an estimation of the uncertainties of severe gusts at observational sites. This could be utilised as a valuable application when forecasting severe winter storm events to determine the possible range of maximum wind gusts and their related losses. This is of relevance for both society and for applications in insurance industry as well. The results for winter storm Kyrill reveal the genesis of a secondary cyclone along the occluded front of the parent cyclone. This is an uncommon location for secondary frontal development and has not been documented in recent review articles covering this field of research. The formation of the secondary cyclone was associated with negative deformation
I Abstract . stretching and supported by diabatic processes in the lower and mid troposphere. The analysis of severe wind gusts associated with the strong cold front over Central Europe reveals the existence of a conditionally instable boundary layer in addition with a turbulent flow. This indicates that high momentum at the top of the boundary layer could have been mixed downward to the ground leading to the strong surface wind gusts. The analyses of winter storm Xynthia show that moist and warm air masses over the anomalously warm North Atlantic Ocean were incorporated into the cyclone. The realisation of sensitivity studies with modified (lowered) sea surface temperatures (SSTs) or reduced surface latent heat fluxes reveal their important influence on the intensity of the storm. A stronger reduction of SST or surfaces fluxes leads to a less intensive cyclone, which emphasizes the importance of warm and moist air near the ocean surface. This is also indicated by reduced diabatic heating rates at lower and mid levels and a weakening of the PV (potential vorticity) tower in case of altered surface conditions. These findings may be of relevance within the context of climate change and possible warming of SSTs. To conclude, the ability of the COSMO-CLM to provide realistic simulations (including realistic area-wide wind gust estimates) of winter storms over the North Atlantic – European sector is ascertained. The realistic representation of near surface wind gusts by the model permits the possibility of estimation of losses and thus is of potential importance e.g. in the insurance business. Furthermore, the outcomes of this thesis extends the current knowledge and provides a substantial basis for the understanding of dynamical aspects and mesoscale mechanisms being relevant during the genesis, development and the passage of individual winter storms like Kyrill (January 2007) and Xynthia (February 2010) over Europe. Finally, the understanding of physical mechanisms and the effects of atmospheric conditions associated with individual winter storms are essential to improve the accuracy of the prediction of future storm events.
II Kurzzusammenfassung .
Kurzzusammenfassung
Extratropische Zyklonen über dem Nordatlantik zählen zu den gefährlichsten und schadensträchtigsten Naturgefahren in Europa. Obwohl der größte Teil der extremen extratropischen Zyklonen in nordöstlicher Richtung an Europa vorbeizieht, sind jedes Jahr Teile Europas von einzelnen starken Stürmen betroffen. Ihre zerstörerische Kraft ist vorrangig andauerndem starken Wind bis hin zu schweren Orkanböen sowie enormen Niederschlagsmengen geschuldet. Insbesondere der Zusammenhang zwischen Sturmböen und resultierenden Schäden ist Gegenstand aktueller Forschung. Aus diesem Grund richtet sich der Fokus dieser Studie auf extreme extratropische Zyklonen über Europa während des Winterhalbjahres (Winterstürme). Detaillierte Untersuchungen dieser Winterstürme hinsichtlich dynamischer Aspekte und mesoskaliger Prozesse während verschiedener Entwicklungsstadien werden mit Hilfe hochaufgelöster Simulationen eines regionalen Klimamodells (COSMO-CLM) durchgeführt. Im ersten Teil dieser Arbeit wird untersucht inwieweit das COSMO-CLM in der Lage ist extreme Winterstürme hinreichend genau wiederzugeben. Zu diesem Zweck wurden insgesamt 158 historische Winterstürme zwischen 1972 und 2008 simuliert. Um flächendeckende Informationen über die räumliche Verteilung der simulierten Böen zu erhalten wurde zusätzlich eine neuartige, um einen probabilistischen Ansatz erweiterte, Böenparametrisierung im COSMO-CLM implementiert. Der zweite Teil dieser Arbeit beschäftigt sich mit der ausführlichen Analyse zweier schadensintensiver Winterstürme der jüngeren Vergangenheit (Kyrill, Januar 2007; Xynthia Februar 2010). Der Fokus liegt hier auf der Betrachtung der dynamischen Aspekte und mesoskaligen Prozesse, die während der Sturmentwicklung eine bedeutende Rolle gespielt haben. Es zeigt sich, dass das COSMO-CLM in der Lage ist die ausgewählten Winterstürme durchweg zufriedenstellend wiederzugeben. Des Weiteren liefert die neue Böenpara- metrisierung realistische und mit anderen Verfahren vergleichbare Resultate. Durch den probabilistischen Ansatz ist zusätzlich eine stationsbezogene Abschätzung der Unsicherheiten der simulierten Böen gegeben. In der Vorhersage ist somit die Möglichkeit gegeben, die Spannweite der zu erwartenden Böen, und somit auch der damit verbundenen Schäden durch Winterstürme, angeben zu können. Die möglichst genaue Vorhersage von Böen ist sowohl
III Kurzzusammenfassung . von gesellschaftlichem Interesse als auch für Anwendungen in der Versicherungsbranche von eindeutiger Relevanz. Die Untersuchungen zu Wintersturm Kyrill zeigen, dass eine sekundäre Zyklogenese entlang der Okklusionsfront des Sturmtiefs stattgefunden hat. Dies ist ein ungewöhnlicher und seltener Fall einer sekundären Entwicklung an Fronten und wird in vorhandenen Übersichtsartikeln zu diesem Thema nicht erwähnt. Die Entstehung der Sekundärzyklone steht in engem Zusammenhang mit negativer Streckungsdeformation entlang der Okklusionsfront sowie diabatischen Prozessen in der unteren und mittleren Troposphäre. Das Auftreten von starken Böen entlang der Kaltfront über Zentraleuropa steht im Zusammenhang mit einer bedingt labilen und turbulenten Grenzschicht. Diese Bedingungen ermöglichen das Heruntermischen hoher Windgeschwindigkeiten vom oberen Rand der Grenzschicht bis hinunter zum Boden. Die Analyse von Wintersturm Xynthia zeigt, dass warme und feuchte Luftmassen über dem ungewöhnlich warmen südöstlichen Nordatlantik an dessen Entwicklung entscheidend beteiligt waren. Unter Berücksichtigung von Sensitivitätsstudien mit verringerter Meeresober- flächentemperatur bzw. reduziertem latenten Wärmefluss kann deren Einfluss auf die Existenz der feucht-warmen Luftmassen und somit auf die Sturmentwicklung quantifiziert werden. Je stärker die Abnahme der Meeresoberflächentemperatur bzw. des latenten Wärmeflusses angenommen wird, desto schwächer ist der resultierende Sturm. Zudem zeigt sich unter modifizierten Bedingungen eine deutliche Abnahme der diabatischen Erwärmungsrate in der unteren und mittleren Troposphäre, was mir einer Abnahme der Mächtigkeit der vertikalen Verteilung der potentiellen Vorticity einhergeht. Die Abhängigkeit der Sturmstärke vom Zustand der Meeresoberfläche ist im Rahmen eines zukünftigen Klimawandels durchaus von Bedeutung. Zusammenfassend lässt sich sagen, dass das COSMO-CLM in der Lage ist Winterstürme (und die damit verbundenen Böenfelder) über dem Nordatlantik und Europa realistisch wiederzugeben. Die Simulation von bodennahen Böen eröffnet die Möglichkeit der Abschätzung von Schäden und bietet somit Anwendungsmöglichkeiten beispielsweise in der Versicherungswirtschaft. Zusätzlich erweitern die Erkenntnisse dieser Arbeit das Verständnis dynamischer Aspekte und mesoskaliger Prozesse, die entscheidend zur Entwicklung von Winterstürmen (Kyrill und Xynthia) beigetragen haben. Ein umfassendes Verständnis der physikalischen Mechanismen und atmosphärischen Randbedingungen, die mit der Entstehung einzelner Winterstürme in Verbindung stehen, ist für die Vorhersage zukünftiger Sturmereignisse von essentieller Bedeutung.
IV
Contents
Abstract ...... I
Kurzzusammenfassung ...... III
Contents ...... V
1. Introduction ...... 1 2. Extratropical cyclones ...... 3 2.1 Brief history of advances on Extratropical cyclones ...... 3 2.2 Winter storms in the Atlantic - European sector ...... 5
3. Winter storm modelling and wind gust estimation with COSMO-CLM ...... 11
4. Case study of winter storm Kyrill (January 2007) ...... 31 5. Case study of winter storm Xynthia (February 2010) ...... 85
6. Summary of the results, discussion and outlook ...... 101
6.1 Paper I ...... 102
6.2 Paper II ...... 103
6.3 Paper III ...... 104
6.4 Discussion and outlook ...... 105
References ...... 109 Acknowledgments ...... 115
Eigene Beteiligung an den Veröffentlichungen ...... 117 Erklärung ...... 119
V
Introduction
1. Introduction
Extratropical cyclones (ETCs) are common everyday meteorological phenomena in the mid-latitudes. Their occurrence is accompanied by rapidly changes of local weather conditions, both in terms of temperature, precipitation and wind. Furthermore, the cyclones themselves are influenced by a variety of environmental conditions that affected their life cycle, path and intensity. The spatial extent and severity of single events puts them among the most costly and dangerous natural hazards in case they affect Europe (e.g. Held et al., 2013). After Mailler et al. (2006) the most damaging European storms belong to one of the following three types: (1) serial storm, which are successive occurring events like Lothar1 and Martin (1999) (Ulbrich et al., 2001) or the storm series in the winter of 1989/1990 (Klawa and Ulbrich, 2003), (2) rapid developers, which exhibit deepening rates exceeding 24 hPa per day (also known as explosive cyclones or “bombs” e.g. Sanders and Gyakum, 1980) like Kyrill (2007) (Fink et al., 2009) or Xynthia (2010) (Liberato et al., 2013) and (3) slow movers, which are able to produce persistent large accumulations of precipitation concentrated over small regions (e.g. Elbe-Flood 2002; Ulbrich et al., 2003, European summer flood 2013; Grams et al., 2014). However, besides their perils, ETCs play a major role in compensation the latitudinal energy imbalance by transporting heat and moisture from the subtropics towards the cold Polar Regions (Oort, 1971). The main intention of this study is to achieve a better understanding of mesoscale processes that play a role on the generation of strong wind gusts and thus on the formation and reorganising of winter storm events that affected Europe in the recent past. Asides from the analysis of a broad range of large-scale atmospheric fields, the realisation of realistic simulations with a non-hydrostatic regional climate model (COSMO-CLM, cf. Rockel et al., 2008) is used to achieve this purpose. Additionally, the representation of a newly physical based wind gust estimation method, extended by a probabilistic approach within the COSMO- CLM is evaluated and compared to already existing wind gust estimation methods. Since wind gust measurements are limited to observation sites, the realistic simulation of area-wide wind gusts during winter storm events provides a strong benefit e.g. for applications in risk assessment. Furthermore, a detailed understanding of the physical mechanisms and the effects
1 Storm names used in this thesis are given as by the Freie Universität Berlin and as used by German Weather Service. Source: http://www.met.fu-berlin.de/adopt-a-vortex/historie 1 Introduction of atmospheric conditions associated with individual winter storm events is essential to improve the accuracy of the prediction of future storm events. To accomplish this aim, the three included publications provide the basis for this thesis by addressing the following current issues:
§ Evaluation of the COSMO-CLM performance and introduction of a novel physical based wind gust estimation method on basis of 158 historical European winter storm events (Paper I).
§ Investigation of dynamic aspects of winter storm Kyrill (2007) producing severe wind gusts over Central Europe in association with secondary cyclogenesis over the eastern North Atlantic (Paper II).
§ Considering the effects of anomalous high SSTs along the cyclone track on the development and intensity of winter storm Xynthia (2010) (Paper III).
Besides the selection criteria due to exceptional process-related characteristics of the individual winter storms Kyrill and Xynthia, the relevance in terms of corresponding losses is considered. Following loss estimates of leading reinsurers, Kyrill was ranked as the 2nd costliest ($10 billion economic losses2 in Europe) winter storm after Lothar ($11.4 billion economic losses) since 1950. With a total economic loss of $6.1 billion, winter storm Xynthia is ranked 4th. These high losses reveal the relevance of the selected winter storms also for society and economy. This thesis is organised in the following way. Chapter 2 gives a short revision of the current state of scientific knowledge on extratropical cyclones. This includes a brief history of advances on the research of ETCs, an overview of ETCs in the Atlantic – European sector and a short introductory survey on wind gusts and their estimation techniques. Chapter 3 – 5 provide the relevant publications (Paper I - III) on which this thesis is based on. A summary and discussion of the main findings of the papers as well as an outlook of possible further work is given in chapter 6.
2 Loss data taken from “Top 10 Losses – Europe; Costliest EU Windstorm/Winter Storm Events” available at: http://catastropheinsight.aonbenfield.com/Pages/Home.aspx 2 Extratropical cyclones
2. Extratropical cyclones
2.1 Brief history of advances on Extratropical cyclones
First efforts in describing the structure and life cycle of extratropical cyclones (ETCs) as a whole have been carried out by Bjerknes and Solberg (1922) and led to the polar front theory of cyclones, also known as Norwegian frontal cyclone model. This conceptual model of the life cycle of an ETC is still widely accepted, although it has been modified several times (e.g. Shapiro and Keyser, 1990, Browning et al., 1994). As the polar front theory was established during the early years of the 20th century, upper air observations were not available (Reed, 1990). The whole cyclone life cycle was deduced from ground-based observations, starting with a wave disturbance along the polar front that separates tropical and polar air masses. Further amplification leads to the typical structure of a frontal cyclone, consisting of a warm sector bounded by a leading warm and a following cold front (Fig. 1a), which both exhibit typical cloud distributions (e.g. Browning and Roberts, 1994). The last stage of the cyclone is associated with the occlusion process (Schultz and Maas, 1993, Schultz and Vaughan, 2011) where the warm air is lifted up together with a shift of the cyclone towards the cold side of the polar front and finally leads to the decay of the cyclone.
IV IV a) b) III III
II II I I
Figure 1. (a) Conceptual model of a Norwegian cyclone showing lower tropospheric (e.g. 850 hPa) geopotential height and fronts for different stages of the cyclone development (caption and figure adapted from Fig. 15a in Schultz and Vaughan, 2011) (b) Conceptual model of frontal-cyclone evolution proposed by Shapiro and Keyser (1990) (Caption and figure adapted from Fig. 2 in Semple, 2003).
During the time of the Second World War, regular upper air observations became more and more frequent, leading to new insights of atmospheric processes. Among them was the discovery of the existence of a westerly flow including embedded so-called Rossby-Waves (after Rossby, 1939), which are of planetary scale (usually 4-6 meanders can be observed
3 Extratropical cyclones along the entire northern hemisphere). Investigations of the effects of velocity of the westerly background flow and propagation speed on the waves led to the relation of upper level divergence/convergence and surface pressure fall/rise by means of the pressure tendency equation (Bjerknes and Holmboe, 1944). Within their research the broadly familiar terms of ‘Trough’ and ‘Ridge’ were created, describing the direction (north- or southbound) of the wave’s amplitude. The introduction of the theory of baroclinic instability (Charney, 1947, Eady, 1949) was an upcoming approach and milestone to describe the occurrence and growth of ETCs. In their independent research, they found out that waves in a baroclinic zone (lapse- rate on isobaric surfaces) are able to become unstable and thus may trigger ETC development. Together with the formulation of the general view on planetary flow patterns in the atmosphere (Rossby, 1940) and the detection of the jet stream (Palmen, 1948), the essentials for outstanding efforts in several branches of research on ETCs were provided. In recent years more and new knowledge has been obtained and many modifications and extensions of the Norwegian model lead to diverse conceptual models for different cyclone development mechanisms (e.g. review paper by Semple, 2003). A schematic overview of the fundamental Norwegian frontal cyclone model and the more recent conceptual model proposed by Shapiro & Keyser (1990) is given in Figure 1. Another, more descriptive concept that describes the three dimensional airflow through an ETC is the principle of conveyor belts (e.g. Carlson, 1980; Browning, 1994; Semple, 2003). These system-relative airflows can be used to describe e.g. the developing cloud structure of an ETC. The two main airflows associated with frontal zones are the warm conveyor belt (WCB, Harrold, 1973) and the cold conveyor belt (CCB, Carlsson, 1980). The WCB forms ahead of the cold front and transports warm and humid air masses poleward from the lower troposphere at its southern end towards the upper troposphere at its northern end. Due to the ascending motion within the WCB, it is accountable for the elongated observed band of clouds along the cold front (Browning, 1986). The CCB originates at low levels ahead of the warm front and moves westward. It undercuts the poleward moving WCB with its associated precipitation and thus redistributes moisture within the system. Furthermore, there is a third type of airflow originating near the tropopause and descending behind the cold front towards the mid troposphere. The so-called dry intrusion (DI, Browning, 1997) is characterised by dry air masses and high values of potential vorticity. The DI can be identified as a cloud-free area (dry slot) in the water vapour, infrared and visible products of satellite imagery. Additionally, the DI is able to create potential instability as it overruns the cold front and thus the warm air associated with the WCB (Browning, 1997).
4 Extratropical cyclones
2.2 Winter storms in the Atlantic - European sector
Extratropical cyclones in the Atlantic - European sector, and particularly winter storms affecting Europe, are a main field of research for a considerable time. This sub-chapter gives a brief overview on the research that has been carried out with focus on the Atlantic - European sector. Most of the North Atlantic ETCs originate as small perturbations at the western parts of the North Atlantic basin, near the warm western oceanic surface currents. This region is also known as the North Atlantic storm track (Hoskins and Valdes, 1990). It is commonly characterised by a strong meridional temperature gradient along the hyper- baroclinic polar front that separates warm subtropical air masses in the south and polar air masses to the north (Pinto et al., 2009). As a result of thermal wind balance, strong baroclinicity is associated with a strong upper-troposheric jet stream located on the warm side of the polar front (Carlson, 1991). This baroclinicity is of essential importance for ETC development (e.g. Hoskins and Hodges, 2002; Gray and Dacre, 2006). Since the upper level jet stream is associated with divergence at the right entrance and left exit region of the jet maximum (Uccellini and Johnson, 1979), it plays a crucial role in enhancing the evolution of ETCs. Baehr et al. (1999) showed that ETCs crossing of the jet stream undergo a rapid deepening phase. Pinto et al. (2009) determined climatologies of the occurrence of extreme
Figure 2. Cyclone track density (cyclone days/winter) of extreme cyclones over the North Atlantic and Europe for NCEP (1958-1998). The tracks (points at 6-hourly intervals, also from derived NCEP) of winter storms Kyrill (blue, starting at 1800 UTC 17 January 2007) and Xynthia (red, 1200 UTC 25 February 2010) are included. (Caption and figure adapted from Fig. 4 in Pinto et al., 2009).
5 Extratropical cyclones and non-extreme ETCs based on NCEP-reanalysis data (Kalnay et al., 1996) for the period 1958-1998. Here, extreme cyclones are classified as the 10% strongest of all identified ETCs. A climatology of the cyclone track density for extreme cyclones during winter together with the tracks of the recent winter storms Kyrill and Xynthia (that are the focuses of Paper II and Paper III) is presented in Figure 2. In general, extreme cyclones (as well as non-extreme cyclones, not shown) tend to move towards the northeast, with only a few systems affecting Europe each winter. The current state of the North Atlantic storm track and thus the tracks of the ETCs are closely related to the phase of the North Atlantic Oscillation (NAO, e.g. Wanner et al., 2001). The NAO (based on the pressure difference between the semi-permanent Icelandic Low and the Azores High) is the leading pattern of variability in the North Atlantic and refers to the redistribution of atmospheric mass between the subtropical Atlantic and the Polar Regions (Hurrell and Deser, 2009). Changes from one NAO phase to another are associated with changes of the direction and strength of the surface westerlies across the North Atlantic towards Europe (Hurrell, 1995). Pinto et al. (2009) figured out that extreme cyclones occur more (less) often during positive (negative) phases of the NAO. Additionally, a shift of the NAO dipole towards Europe during positive phases results in an enhanced background pressure gradient that favours cyclone activity over Europe e.g. in the case of winter storm Kyrill (2007) (Fink et al., 2009). Nevertheless, even during negative NAO phases extreme cyclones can occur and affect Europe as in the recent case of winter storm Xynthia (2010). A negative NAO phase is usually associated with a southward shift of the upper level jet stream (Woolings et al., 2010), which plays a crucial role on the far southern formation of Xynthia. Besides the well-known region of ETC occurrence outlined above, Ayrault et al. (1995) detected a distinct area downstream and slightly south of the climatological storm track location where frontal waves are able to develop during zonal weather regimes. These secondary cyclones often originate along the intensive trailing cold front of a parent cyclone and can have large growth rates (Parker, 1998). For example, winter storm Kyrill was identified as an unusual case of secondary cyclogenesis as the secondary cyclone developed at the occluded front of the parent cyclone (see Paper II of this thesis). As summarised by Parker (1998), various processes are important for the growth of frontal waves. These processes include shear at the frontal zone (e.g. Joly and Thorpe, 1991), large-scale strain (e.g. Renfrew et al., 1997), latent heat release (e.g. Hoskins and Berrisford, 1988; Ahmadi-Givi et al., 2003), boundary layer processes (Adamson et al., 2006) and the influence of a local stripe of
6 Extratropical cyclones maximum boundary layer potential vorticity (PV) that is associated with barotropic instability (cf. Figure 1 in Dacre and Gray, 2006). Besides the effects of high low-level PV on secondary cyclogenesis, the PV-concept first used by Rossby (1940) and Ertel (1942) and enhanced by Hoskins et al. (1985) can be used to explain and analyse the evolution of ETCs. The two basic properties of PV are (1) conservation (PV is conserved in case of adiabatic motion) and (2) invertibility (under suitable balance conditions, such as geostrophic balance, the wind and temperature field can be derived from PV if it is given everywhere) (Hoskins, 1997). PV is defined as:
1 PV = ζ ⋅∇θ ρ where ρ is the density, ζ the absolute vorticity and ∇ θ the gradient of the potential temperature. PV is often expressed in PV units (1 PVU = 10-6 m2 s-1 K kg-1). A potential application of this PV concept is for instance the definition of the dynamic tropopause (Hoskins, 1990). While the climatological distribution of PV has values between 0 and 1 PVU within the troposphere, there is a sharp increase of the static stability between the upper troposphere and lower stratosphere that leads to enhanced PV values, where the 2 PVU surface corresponds to the dynamic tropopause. The PV concept also allows for explaining cyclogenesis in case that a positive upper-level PV anomaly arrives over a low-level baroclinic region (Hoskins, 1985). Figure 3 illustrated the interaction of such an upper level PV anomaly and the induced circulation. The positive upper-level PV anomaly is associated with cyclonic circulation and induces a cyclonic circulation that extends through the troposphere down to the surface (Fig. 3 a). The low-level circulation in turn creates a low- level positive temperature anomaly by advection of warm air towards the north and somewhat ahead of the upper-level PV anomaly (Fig. 3 b). This warm low-level temperature anomaly
Figure 3. Schematic overview of cyclogenesis associated with the arrival of an upper-level PV anomaly over a low-level baroclinic region (Caption and figure adapted from Fig. 21 in Hoskins, 1985). See text for details.
7 Extratropical cyclones again induces a cyclonic circulation and thus is able to reinforce the upper-level circulation pattern. As mentioned before, PV is a conserved quantity in case of adiabatic motions. Since diabatic processes considerably determine the development of extratropical cyclones, the application of the PV perspective is a valuable tool to analyse processes that are important for the evolution of ETCs. Many studies have emphasised the importance of diabatic processes, in particular latent heat release due to condensation of water vapour, on the development of ETCs (e.g. Danard, 1964; Tracton, 1973; Uccellini, 1990). In a recent study by Fink et al. (2012) the role of diabatic processes on the development of five recent winter storms (Lothar and Martin (1999), Kyrill (2007), Klaus (2009, cf. Liberato et al., 2011) and Xynthia (2010)) is quantified. The main finding of this study is that the pressure fall of three investigated storms (Lothar, Klaus and Xynthia) is mainly related to diabatic processes, while baroclinic processes are dominant for Martin and Kyrill. The key effect of latent heat release by condensation is the generation of anomalously high PV in the lower und mid troposphere (e.g. Reed et al., 1992). If these high PV anomalies interact with a positive upper-level PV anomaly, they form a so-called PV tower (Rossa et al., 2000) that extends vertically through the troposphere and often is associated with rapidly deepening cyclones like the “October Storm” (Hoskins and Berrisford, 1988), the “Presidents Day Cyclone” (Whitaker et al., 1988), winter storm Lothar (Wernli et al., 2002) or the more recent winter storm Xynthia (Campa and Wernli, 2012; also see Paper III of this thesis). Furthermore, diabatic processes, like they occur in WCBs, are able to modify the upper-tropospheric wave guide (Grams et al., 2011). Along the airstream of a WCB, the PV increases due to condensational heating as long as the air parcels are below the level of maximum diabatic heating (see Figure 4 b in Wernli and Davies, 1997). During further ascent close to the tropopause region, a reduction of PV occurs, leading to negative upper-level PV anomalies. These negative PV anomalies in turn can have a significant impact on the downstream flow evolution like the formation of meridional elongated PV-streamers. Massacand et al. (2001) identified an upper level PV-streamer as a precursor for a high-impact weather of a Mediterranean cyclonic development. Likewise, the evolution of winter storm Xynthia is associated with an upper-level PV streamer (Piaget, 2011, also see Paper II of this thesis) Another widespread field of research associated with ETCs covers the connection between intense winter storms and the extremes in surface winds. These extremes in surface winds are often associated with convective downdrafts along cold fronts (e.g. Houze and Hobbs, 1982) or convective systems like derechos (e.g. Gatzen et al., 2011). Also for non-
8 Extratropical cyclones convective high winds, like they occur in an environment of steep pressure gradients or within low-level jets (Browning and Pardoe, 1973), some detailed further physical explanations exist (Knox et al., 2011). For example, tropopause folds (e.g. Uccellini, 1990) could be related to high surface winds (Browning and Reynolds, 1994). The authors figured out that during a severe wind event in the UK 1991, high-momentum stratospheric air descends to the boundary layer, and then was transferred to the surface via shear instabilities (Knox et al., 2011). More recently, the sting jet hypothesis (first set up by Grønas, 1995) became more and more established since for different severe storm events high surface wind speeds could be linked to a sting jet (e.g. Great Storm over UK in October 1987 (Browning, 2004; Clark et al., 2005), winter storm Jeanette in October 2002 (Parton et al. 2009) or winter storm Gudrun in January 2005 (Baker, 2009)). Sting jets evolve at the hooked tip of the cloud head that forms when the bent-back warm front and the CCB wrapped around the low centre (cf. Fig. 1 in Baker, 2009). So far, cyclones that have been associated with sting jets show a similar structure and development corresponding to the conceptual model by Shapiro and Keyser (1990) (Baker, 2009). In the presence of multiple mesoscale slantwise circulations (that may have been caused by conditional symmetric instability (CSI, Schultz and Schumacher, 1999)), air may leave the tip of the cloud head and enters the dry slot below where rapid evaporation and diabatic cooling causes further downward acceleration immediately upwind of the area of damaging surface winds (Browning, 2004). Although the mechanisms leading to severe wind gusts are generally understood, their determination by means of atmospheric models is still a challenging issue. In particular, the proper estimation of losses requires a realistic representation of area-wide wind gusts. Klawa and Ulbrich (2003) derived a relationship between wind speed above a certain threshold and the estimation of losses that corresponds to the proportionality
loss ~ (maximum wind speed)3.
This implies that during high-wind situations, relatively small increases in wind speed can have a disproportionate impact on the amount of wind damage (Browning, 2004). Since the climatology of wind gusts does not coincides with the climatology of mean wind, a simple relation between mean wind and gust cannot be derived (Brasseur, 2001). For that purpose, a variety of wind gust estimation methods have been developed and applied to atmospheric models to obtain realistic area-wide distributions of wind gusts and/or associated losses during severe weather events for both present and future climate conditions (e.g. De Rooy and Kok, 2004; Della-Marta et al., 2010; Pinto et al., 2010; Schwierz, 2010; Etienne et al., 2013).
9 Extratropical cyclones
In a first approach, Durst (1960) uses a gust factor derived as the fraction between wind gusts and mean wind speed to predict gusts. This technique has been refined later to take into account the state of the atmosphere in terms of stability or the roughness length in the environment (e.g. Wieringa, 1973; Verkaik, 2000). In an approach by Brasseur (2001), wind gusts are interpreted as downward transition of high-level boundary-layer momentum in case that turbulent kinetic energy (TKE) is able to overcome buoyancy force. Finally, the understanding of gusts as a combination of mean wind speed amplified by a part that can be connected with TKE (see Paper I of this thesis) should be mentioned as an alternative to predict wind gusts. In case that TKE is not directly available, Schulz and Heise (2003) make use of friction velocity as a predictor for turbulence.
10 Paper I
3. Winter storm modelling and wind gust estimation with
COSMO-CLM
Journal article (published):
BORN, K., P. LUDWIG, AND J. G. PINTO, 2012: Wind Gust Estimation for Mid-European Winter Storms: Towards a Probabilistic View. Tellus A 64:17471 doi: 10.3402/tellusa.v64i0.17471
Permission to reprint:
The article and any associated published material is distributed under the Creative Commons Attribution 3.0. Copyright on this article is retained by the authors.
Original page numbers of the manuscript are used.
11
SERIES A DYNAMIC METEOROLOGY AND OCEANOGRAPHY PUBLISHED BY THE INTERNATIONAL METEOROLOGICAL INSTITUTE IN STOCKHOLM
Wind gust estimation for Mid-European winter storms: towards a probabilistic view
By KAI BORN*, PATRICK LUDWIG and JOAQUIM G. PINTO, Institute for Geophysics and Meteorology, University of Cologne, Kerpener Str. 13, 50937, Cologne, Germany
(Manuscriptreceived 25 May 2011; in final form 9 January 2012)
ABSTRACT Three wind gust estimation (WGE) methods implemented in the numerical weather prediction (NWP) model COSMO-CLM are evaluated with respect to their forecast quality using skill scores. Two methods estimate gusts locally from mean wind speed and the turbulence state of the atmosphere, while the third one considers the mixing-down of high momentum within the planetary boundary layer (WGE Brasseur). One hundred and fifty-eight windstorms from the last four decades are simulated and results are compared with gust observations at 37 stations in Germany. Skill scores reveal that the local WGE methods show an overall better behaviour, whilst WGE Brasseur performs less well except for mountain regions. The here introduced WGE turbulent kinetic energy (TKE) permits a probabilistic interpretation using statistical characteristics of gusts at observational sites for an assessment of uncertainty. The WGE TKE formulation has the advantage of a ‘native’ interpretation of wind gusts as result of local appearance of TKE. The inclusion of a probabilistic WGE TKE approach in NWP models has, thus, several advantages over other methods, as it has the potential for an estimation of uncertainties of gusts at observational sites. Keywords: windstorm, wind gust estimation, TKE, COSMO-CLM, probabilistic approach
1. Introduction Schwierz et al., 2010). In these studies, very different approaches for wind gust estimation (WGE) are used. Wind gusts associated with windstorms are one of the main From this fact, the following questions arise: Which sources of economic and insured losses over Europe. For complexity of a WGE approach is necessary to obtain example, storm Kyrill (18 January 2007) caused insured good WGEs? Which numerical weather prediction (NWP) losses of about t2.4 billion in Germany alone and caused a model information may be provided that contributes to a widespread disruption of normal social activities, public WGE? Is a simple and self-suggesting approach based on transportation and energy supply, as well as a large number the definition of subscale kinetic energy able to consider the of fatalities over large parts of Europe (cf. Fink et al., obvious stochastic nature of gusts, and how does it 2009). Therefore, the correct estimation and forecast of compare to standard WGE methods? wind gusts associated with winter storms may enhance the Simulated near-surface winds from NWP models are capability of issuing accurate severe weather warnings and usually smaller than observed wind gusts. This fact is related is of great value in scientific, societal and economical terms. to (1) the formulation of model variables as averages over a Several studies on the estimation of gusts associated with space and time (grid box and time step) and (2) the high the passage of windstorms were recently undertaken either temporal variability of gustiness, especially during strong using mesoscale modelling or statistical approaches (e.g. wind episodes. From the observational point of view, gust Brasseur, 2001; Goyette et al., 2003; De Rooy and Kok, parameterisation reduces to the problem how a probability 2004; Agustsson and Olafsson, 2004, 2009; Friederichs et distribution of highly resolved wind speeds changes when al., 2009; Pinto et al., 2009). One of the recent applications the according time series is averaged. For NWP applica- is to estimate potential losses associated with wind gusts tions, model-resolved variables like wind speed and (e.g. Della-Marta et al., 2009, 2010; Pinto et al., 2010; measures for the state of turbulence can be used to estimate gusts. In general, three techniques have been established: (1) *Corresponding author. the use of a gust factor as fraction between gust and mean email: [email protected] wind speed (based on the original work of Durst, 1960;
Tellus A 2012. # 2012 K. Born et al. This is an Open Access article distributed under the terms of the Creative Commons Attribution-Noncommercial 3.0 1 Unported License (http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited. Citation: Tellus A 2012, 64, 17471, DOI: 10.3402/tellusa.v64i0.17471
(page number not for citation purpose) 2 K. BORN ET AL.
e.g. Wieringa, 1973; Verkaik, 2000), varying with historical European windstorms is considered. These were atmospheric stability and/or roughness length in the envir- simulated by means of the regional climate model, onment; (2) the interpretation of gusts as downwards- COSMO-CLM, using reanalysis data as boundary transition of higher level boundary-layer momentum (e.g. conditions. Brasseur, 2001; Brasseur etal., 2002) and (3) theunder- This study is organised as follows: Section 2 describes standing of gusts as mean wind plus a part connected with data and the NWP model, while Section 3 presents the turbulent kinetic energy (TKE). If TKE is not available, different WGE formulations used. The evaluation of WGE wind drag in terms of friction velocity (e.g. Schulz and Heise, methods (Section 4) is divided into four steps: (1) analysis 2003), atmospheric stability indices and wind direction, of statistical characteristics of observational data, (2) an describing the advection of TKE from near-by regions overall evaluation of COSMO-CLM simulations, (3) an with different roughness characteristics (Agustsson and exemplary comparison of WGE for typical winter storm Olafsson, 2004), can be used as a proxy for the turbulence events and (4) the calculation of skill scores for all state. events. The discussion of the results is presented in Section Wind gusts are affected by particular characteristics of 5, and a short summary and conclusion finishes this study the model topography, mainly land cover (in terms of (Section 6). roughness) and surface elevations, which induce turbulent eddies and, thus, influence the turbulence state of the atmosphere. The WGE formulation has to consider this 2. NWPmodel and data subscale influence; its quality depends on the calibration of As a basis for this study, model simulations of 158 historical turbulence-related WGE parameters. One major setback is European windstorms between 1972 and 2008 (see Fig. 1a) that spatially distributed observations usually do not have been undertaken using the mesoscale atmospheric provide sufficient information about the atmospheric model, COSMO (http://www.cosmo-model.org). It is turbulence; a statistical calibration of the turbulence- mainly designed for application on the meso-b/g scale using related part of a WGE is not possible. From the viewpoint grid resolutions from 20 km down to 1 km. The COSMO of atmospheric modelling, wind gusts show a stochastic model has been widely used for regional climate simulations behaviour. Thus, rather than predicting absolute values, (e.g. Bo¨ hm etal., 2008; Jaeger etal., 2008; Rockel etal., the estimation of a range of probability at which a gust 2008; Lautenschlager et al., 2009; see also COSMO-CLM value may occur appears to be an appropriate and skilful community at http://www.clm-community.eu). information. Further, such a probability range is also very In the COSMO model, the non-hydrostatic, fully com- helpful for various applications, for example, when decid- pressible Navier Stokes equations are solved on an Ara- ing whether issuing severe weather warnings (e.g. Wichers kawa-C grid using a hybrid terrain-following coordinate. Schreur and Geertsema, 2008). In the vertical, the model contains the whole troposphere In the following sections, a basic formulation of a and parts of the lower stratosphere, the latter mainly as a turbulence-driven WGE method, hereafter called WGE damping layer. Standard vertical resolutions use 20 45 TKE, considering a probabilistic extension, is described. layers. Physical parameterisations consider an extended Results of two standard WGE methods considering the version of the level 2.5 scheme after Mellor and Yamada turbulence state of the atmosphere locally and non-locally (1982) using prognostic TKE. Cloud microphysics are are compared with this new WGE method. The two based on a Kessler-type scheme but contain cloud ice, standard WGE methods are the German Weather Service graupel, and consider advection of cloud water/ice and (DWD; Deutscher Wetterdienst) approach in COSMO- rain/snow. Radiation effects are estimated using the d-two- CLM, which uses friction velocity as predictor for turbu- stream approximation (Ritter and Geleyn, 1992). The lence (Schulz and Heise, 2003; Schulz, 2008), and the model has been developed by the DWD and is in approach of Brasseur (2001), which estimates gusts con- operational use for regional NWP in several European sidering a possible downward transition of air from higher weather services. More detailed information may be found atmospheric levels, carrying high momentum. The new in Steppeler et al. (2003). WGE TKE approach defines the maximum available In this study, COSMO was used in its climate version kinetic energy by interpreting TKE in a statistical sense COSMO-CLM4.0 (Bo¨ hm etal., 2008). The mostimportant as measure for wind speed variance. The probabilistic difference to the NWP version is that no assimilation of extension assesses the probability range of local gust observational data and no nudging have been applied. In factors statistically from observations. The forecast cap- the vertical, 32 layers in the hybrid pressure-based terrain- ability of the methods is tested by computation of proper following coordinate are used; the horizontal grid consists skill scores. For the evaluation of WGE methods, a set of of 257 271 grid boxes with grid sizes of 0.1658 resolution WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 3
Fig. 1. (a) Year and month of simulated storms from 1972 to 2008, in a total of 158 storms and (b) COSMO-CLM model region, including orography, colour scale in m. 4 K. BORN ET AL.
on a rotated latitude longitude grid centred on 88W, displacement dz between surface heights at observational 50.758N. The thickness of the lowest model layer is sites and the average model grid box height is considered by approximately 67 m. The first full level, where horizontal adding a correction factor ð@vmax=@zÞÁ@z. This kind of momentum and temperature is calculated, is, thus, roughly first-order correction is absolutely necessary for a compar- at33.5 m above ground. The Runge Kutta integration ison between grid box averages of model simulations and scheme with a time step of 90 s and an output interval of local observations. 1 h was used. In general, the simulation periods are 96 h, Wind observations are provided for 37 DWD sites and starting 48 h before the peak of the event. For some cases cover the period from 1950 to 2005 (see Table 1). They (e.g. Lothar and Martin), the initialisation time had to be consistof hourly wind records from 1979 to2005, most slightly changed to guarantee a good representation of that observations start in 1976 with 3-hourly reports. The data particular storm. The model domain comprises entire is searched for inhomogeneities; obviously wrong observa- Europe and parts of Northern Africa (Fig. 1b). In this study, tions are omitted (e.g. 50 m s 1 limited maximum winds). we focus on Germany for evaluation of the simulations. Except for mountain sites, the available number of gust In long-term transient COSMO-CLM simulations for observations typically decreases with distance to the coast: Europe (e.g. Bo¨ hm etal., 2008; Jaeger etal., 2008), the This is due to the fact that in Germany gusts are only representation of extreme events like windstorms may reported when they exceed a threshold of 12 m s 1. Such differ considerably from the real event. This fact is due to high gustvalues are less frequentinland. the boundary-only forcing, as atmospheric conditions are For the evaluation of the RCM simulations, a dataset, mainly inferred over the lateral boundaries. For a more including complete life cycles of cyclones obtained from accurate simulation of storms, a shorter model spin-up ERA-Interim, is considered. Each track includes informa- between initialisation and storm formation is advanta- tion (e.g. core pressure, vorticity) for one cyclone at each geous, as it allows for an evolution of the event closer to the time step. The cyclone tracks are computed using an observed development. Therefore, the present set of algorithm originally developed by Murray and Simmonds COSMO-CLM simulations of historical storm events for (1991), which is adapted and evaluated for Northern Germany has been produced. As boundary forcing, ERA40 Hemisphere cyclone properties and high-resolution data- and ERA-Interim reanalyses (Uppala et al., 2005; Dee sets (Pinto et al., 2005; Nissen et al., 2010). Further details et al., 2011) are used. The storms in the overlapping period, on the method, its settings and cyclone climatologies can be 1989 2002, have been simulated using both ERA40 and found in Murray and Simmonds (1991), Simmonds etal. ERA-Interim in order to assess the influence of the change (1999) and Pinto et al. (2005, 2007b). of boundary forcing. It turned out that storm simulations using either ERA40 or ERA-Interim as atmospheric 3. WGE estimation with different formulations forcing do not exhibit systematic differences (not shown); hence, they can rather be understood as different realisa- Wind gust estimation in NWP is a purely diagnostic tion of the same storm. calculation. The model variables are not influenced by The simulated episodes include all major storms, which the WGE. A WGE formulation considering model-pre- affected central Europe between 1972 and 2008. These dicted TKE and a probabilistic estimate of an uncertainty events were selected based on a storm intensity index, range is introduced here. The TKE approach is based on which considers exceedances of the 98th wind speed the relation between mean TKE q and gusts vmax, which can percentiles and is applied to the reanalysis dataset (Klawa be summarised in the relationship: and Ulbrich, 2003; Pintoetal., 2007a; Fink etal., 2009). In pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffi this way, the majority of the top-ranking events of the last vmax ¼ 2Emax ¼ v þ 2q þ ev (1) decades for Germany are collected in the storm list. In g addition, a few weaker events known from insurance or, in a formulation of the gust factor v, which is simply companies’ reports were included. In order to allow for a the ratio gust/mean wind speed: pffiffiffiffiffi comparison with observations, COSMO-CLM output had 2q to be post-processed: In a horizontal plane, the 0.1658 gv ¼ 1 þ þ eg (2) v gridbox averages were interpolated to locations of the observational sites by means of a distance-weighted inter- Here, Emax is the maximum kinetic energy, and ov is the polation using a Gaussian filter (using 9 9 neighbour ‘stochastic’ subgrid-scale part of vmax. The random term grid points, and 0.338 lat/lon 1/e-width), including the eg ¼ ev= v is related to the difference between actual subscale vertical near-surface wind gradients calculated from model kinetic energy of the gust and mean TKE and is of stochastic 10 m winds. The vertical gradients are needed for a height nature for the grid-scale model. It represents the variability correction of winds and gusts: The effect of the vertical of gusts due to the ‘unknown’ portion of small-scale WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 5
Table 1. Information on the 37 observational sites, including WMO number, station name, geographical location and height above seal level
WMO Name Lat Lon Elevation Daily hourly data available from Until Hourly Gusts no. (8N) (8E) (m a.s.l.) values (%)
10020 SYLT 55.01 8.25 26 1 January 1976 1 January 1979 31 December 2005 226933 20.51 10113 NORDERNEY 53.43 7.09 11 1 January 1976 1 January 1979 31 December 2005 202600 15.44 10129 BREMERHAVEN 53.32 8.35 7 1 January 1976 1 January 1979 31 December 2005 231334 8.77 10147 HAMBURG- 53.38 9.59 11 1 January 1976 1 January 1981 31 December 2005 204360 6.33 FUHLS. 10161 BOLTENHAGEN 54.00 11.12 15 1 January 1976 29 August1977 31 December 2005 230103 11.10 10162 SCHWERIN 53.39 11.23 59 1 January 1976 29 August1977 31 December 2005 226453 6.27 10170 ROSTOCK- 54.11 12.05 4 1 January 1976 29 August1977 31 December 2005 229428 9.93 WARNEM. 10224 BREMEN 53.03 8.48 5 1 January 1976 1 January 1981 31 December 2005 211507 6.55 10270 NEURUPPIN 52.54 12.49 38 1 July 1975 29 August1977 31 December 2000 171210 3.73 10291 ANGERMUENDE 53.02 14.00 54 1 July 1975 29 August1977 31 December 2000 170598 5.27 10317 OSNABRUECK 52.15 8.03 95 1 January 1976 1 January 1979 31 December 2005 177067 6.20 10338 HANNOVER- 52.28 9.41 59 1 January 1976 1 January 1976 31 December 2005 238068 4.97 LANG. 10368 WIESENBURG 52.07 12.28 187 11 June 1990 11 June 1990 31 December 2000 88271 10.95 10382 BERLIN-TEGEL 52.34 13.19 36 2 January 1961 2 January 1961 31 December 2000 190182 2.66 10384 BERLIN-TEMP. 52.28 13.24 49 1 January 1950 1 January 1950 31 December 2000 368920 2.11 10385 BERLIN-SCHOEN. 52.23 13.32 45 1 July 1975 29 August1977 31 December 2000 186349 3.61 10393 LINDENBERG 52.13 14.07 98 1 July 1975 29 August1977 31 December 2000 182616 3.86 10396 MANSCHNOW 52.33 14.33 12 11 June 1990 11 June 1990 31 December 2000 85474 9.00 10438 KASSEL 51.18 9.27 231 1 January 1976 1 January 1979 31 December 2005 205664 3.03 10453 BROCKEN 51.48 10.37 1142 1 January 1976 29 August1977 31 December 2005 235536 30.96 10469 LEIPZIG 51.26 12.14 131 1 January 1976 29 August1977 31 December 2005 229602 5.02 10488 DRESDEN 51.08 13.45 227 1 January 1976 29 August1977 31 December 2005 223602 5.27 10499 GOERLITZ 51.10 14.57 238 1 January 1976 29 August1977 31 December 2005 218607 7.75 10513 KOELN-WAHN 50.52 7.10 92 1 January 1976 1 January 1981 31 December 2005 205973 3.16 10609 TRIER- 49.45 6.40 265 1 January 1976 1 January 1979 31 December 2005 211573 5.71 PETRISBERG 10637 FRANKFURT/M. 50.03 8.36 112 1 January 1976 1 January 1981 31 December 2005 202463 4.47 10685 HOF-HOHENSAAS 50.19 11.53 567 1 January 1976 1 January 1979 31 December 2005 220504 7.23 10727 KARLSRUHE 49.02 8.22 112 1 January 1976 1 January 1979 31 December 2005 182201 5.87 10729 MANN HElM 49.31 8.33 96 1 January 1976 1 January 1979 31 December 2005 197812 2.55 10738 STUTTGART-ECH. 48.41 9.14 371 1 January 1976 1 January 1981 31 December 2005 181479 3.23 10763 NUERNBERG- 49.30 11.03 314 1 January 1976 1 January 1981 31 December 2005 194281 2.74 KRA. 10803 FREIBURGI.BR. 48.00 7.51 269 1 January 1976 1 January 1979 31 December 2005 211339 5.19 10838 ULM 48.23 9.57 571 1 January 1976 1 January 1979 31 December 2005 172506 2.60 10852 AUGSBURG- 48.26 10.57 462 1 January 1976 1 January 1979 31 December 2005 198728 3.68 MUEHLH. 10908 FELDBERG/SCHW. 47.53 8.00 1486 1 January 1976 1 January 1979 31 December 2005 211775 24.58 10961 ZUGSPITZE 47.25 10.59 2960 1 January 1976 1 January 1979 31 December 2005 209701 26.26 10980 WENDELSTEIN 47.42 12.01 1832 1 January 1976 1 January 1979 31 December 2005 193686 23.55
In addition, the start/end dates, since/until daily and hourly observations are available. Last two columns mention the amount of available hourly values and the fraction of gusts therein, respectively.
pffiffiffiffiffi kinetic energy. og is not necessarily normally distributed but is the ‘average turbulent wind speed’ vturb ¼ v þ 2q , which has obviously an expected value of 0. In this study, represents the median of the estimated gust distribution. stochastic features of og are derived from observational The derivation of eqs. (1) and (2) and quantile regression data by quantile regression. The model scale parameter used details are shown in Appendixes A.1 and A.2. 6 K. BORN ET AL.
In COSMO-CLM, the standard method for estimating The evaluation of the WGE methods is then undertaken non-convective gusts is to use wind speed interpolated from using proper skill scores. Three scores compare different the lowest model level to 30 m height and the friction characteristics of the WGEs: The correlation (CORR) of velocity u*: time series evaluates accordance of temporal variability, the root mean square skill score (RSS) the deviation from v v 3:0 2:4 u (3) gust ¼ jjz¼30m þ Á Á Ã WGEs to observations, and the quantile skill score (QSS) the similarity of probability distributions of WGEs in terms The maximum gust v is then defined as the maximum max of the quantile functions. Formulae for the skill scores are occurring in an output time interval, which here is 1 h. The listed in Appendix A.3. factors 3.0 and 2.4 are motivated by Prandtl-layer theory (Panofsky and Dutton, 1984); the numerical values are determined empirically. A more detailed description and 4. Results evaluation of this formulation can be found in Schulz and Heise (2003) and Schulz (2008). In general, the friction 4.1. Statistical evaluation of observational data velocity method and TKE approach are relatively similar, In a first step, the relation between observed gusts and because in both cases a predictor for local turbulence is average wind speeds for the observational dataset is estimated; in case of WGE DWD, an empirical factor analysed. In particular, the possible use of multiple linear allows for the optimum adaption to observations. In case regression (MLR) models for spatial interpolation of of WGE TKE, assumptions on the behaviour of the statistical characteristics of gust factors is briefly discussed. stochastic part o have to be made. In this study, the v For this purpose, the Gauss-filtered density of observa- characteristics of o are based on gustobservations. v tions in the (v v )-space using 1/e-filter-widths of Differentfrom theseapproaches, as itdoes notconsider 10m max 2ms 1 in each direction was calculated. Fig. 2 shows the local turbulence directly, is the WGE approach named density plots of wind gusts against mean wind speeds and after Brasseur (2001), henceforth referred to as WGE gust factors for three exemplary sites: one representative of Brasseur. It has been applied in many cases (e.g. Goyette an exposed mountain region, one for a coastal area and one et al., 2003; Pinto et al., 2009) and uses a relation between for a low-range hilly region far from the coast. In addition, buoyancy and TKE in order to decide whether a parcel of quantile regression lines based on a Weibull-like behaviour air may be mixed down from a certain height to the surface, of the distribution of gust factors, dependent on wind speed carrying momentum available for the peak gusts. The basic above a certain level and assuming an exponential power- relation is v ¼ maxðvð^zÞÞ for all levels ^z, where max law relation between average wind speed and gusts Z^z Z^z (see Appendix A.2), have been added to the diagrams. 1 h z0 h z 0 0 vð ÞÀ vð l Þ 0 The medians of the gust factors vary only little as a q ðz Þdz gN dz (4) ^z À zs hvðzl Þ function of wind speed, showing very weak negative slopes zs zs in all cases. This behaviour may be attributed to the fact is satisfied. The inequation questions if the mean TKE, that strong wind conditions lead to near-neutral stratifica- integrated from a near-surface layer zs to a certain height ^z, tion with less variable TKE/wind speed relations. While the is able to overcome the buoyancy in the same air column. median of the gust factors is relatively similar for different Buoyancy is calculated using the deviation of potential locations, the spread of the gust factors’ distribution at virtual temperature uv in the considered height from the constant wind speed is obviously very variable: The width near-surface value and the gravity acceleration gN. Here, zl of the distributions of gust factors depends strongly on is the next lower model level. It has to be noted that in some wind speed, and itincreases withdecreasing mean wind studies, zl is taken as near-surface level (e.g. Goyette et al., speed (see Fig. 2). 2003; Pinto et al., 2009). An upper bounding value is In Fig. 3, the spatial variation of the estimated mean gust formulated, allowing the wind velocity to be taken from the factors is depicted for the observational sites. For this planetary boundary layer (PBL) only. The upper limit is graphic, 10 sites with low counts of gust observations were represented by a dynamic PBL height assumption: PBL excluded. The dependence of mean gustfactors given as height is defined as the vertical level, where TKE is 1% of quantiles from latitude and height are shown as graphs, a the surface TKE. Further, the method considers a lower map of Central Europe shows the location of the sites and bound, which takes into account only the TKE production corresponding average median values. The box and whis- due to vertical movements (see Brasseur, 2001, for more kers, showing 5, 25, 50, 75 and 95% quantiles (q05, q25, details). The mixing approach can be understood as a kind q50, q75 and q95), give an idea of the width of the gust of non-local approach by interpreting the vertical turbu- factors distribution. The first conclusion apparent from the lence structure. data is that there is no clear relation of the gust factor or its WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 7
Fig. 2. Density plots of gust versus 10 m wind speed (upper row) and gust factors versus 10 m wind speed (lower row) for three exemplary climate observation sites, representative for an exposed mid-range mountain (Brocken, 10453), a maritime/coastal region (List/Sylt, 10020) and a low-range hilly region far from the coast (Trier, 10609). Colour shades represent normalised density of observations, lines represent a quantile regression of the gust factors for the 5, 25, 50, 75 and 95% quantiles. For more details on each station, see Table 1.
spread with latitude or elevation of observational sites. as predictors gains with a coefficient of determination of Extremely exposed mountain observations (10453 and 13%, again not a promising result for a potential predictive 10908) are connected with rather small gust factors. This skill of a statistical spatial interpolation. More interesting may be primarily attributed to the fact that in the free than a gust factor itself may be the spread, which is formed atmosphere, weaker turbulence is connected with higher by the difference between q95 and q05. This is a direct average wind speeds. As it would be useful to relate the gust measure for the width of the gust factors distribution and for factors with external parameters of the land cover, linear the uncertainty at which a gust factor can be estimated, models between the median gust factor and potential which may be associated with local topographic character- predictors were tested. Only those parameters that reveal istics. In order to test for the predictability of the (q95 q05)- at least a weak relationship are depicted in Fig. 3, namely, spread, a second multilinear model has been tested. It uses the location and the height of observational sites. A slight the difference of quantiles (q95 q05) as predictand and increase of gust factors with increasing distance to the coast distance from the German Bight, height, roughness length from 1.45 to 1.65 may be observed in Fig. 3a. This increase (z0) and orographic variance within a circle of 10 km is statistically significant at the 95% level (after student’s diameter as predictors (Fig. 4). The topographic character- t-test), but the explained variance is only 11%. A multilinear istics were derived from USGS GTOPO30 and USGS model using height of observational sites and their location Global Land Cover Characterisation 1 km land cover 8 K. BORN ET AL.
Fig. 3. (a) Mean gust factors at observational sites (x-axis) against latitude. (b) Mean gust factors against heights of observational sites. The box and whiskers show mean values for the 5, 25, 50, 75 and 95% quantile, respectively. (c) Mean 50% quantiles of the gust factors are depicted as colour dots on their geographical location. Ten stations with very low numbers of observations have been excluded from this plot. For more details on each station, see Table 1. database. For this purely statistical model, a coefficient of with wind speed obtained for the specific sites, where a determination (COFD, comparable to explained variance) comparison of gusts is intended, provides more appropriate of roughly 33% could be reached (Fig. 4a, topmost row). information than classical empirical gust estimation. This is The predictability is higher than for the gust factor itself, but further discussed in Section 4.4, Fig. 8. for a possible spatial interpolation the results are not convincing, indicating that such a statistical method needs 4.2. Overall evaluation of COSMO-CLM storm improvement. Interestingly, roughness plays only a minor simulations role for the predictive skill. It has to be concluded that the gust factor seems to be First, the performance of the COSMO-CLM storm simula- strongly connected with dynamical features like wind speed tions is discussed by comparing the paths of the storms in or TKE, which have to be taken from model simulations. the RCM simulations with tracks derived directly from Still, an important result from Figs. 2 and 3 is that, in a first ERA-Interim data (see Section 2). Although ERA-Interim order approximation, the consideration of probabilities by has a lower resolution, tracks of the storms obtained from using quantile regression parameters of the gust factors these data are the best available estimate of storm positions WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 9
Fig. 4. Evaluation of the MLR model for the width of local gust factor distributions. Predictors are distance from the German Bight (dist), height of the site (height), roughness length at the site (z0) and orographic variance within a circle of 10 km diameter (oro_var). (a) Adjusted coefficient of determination (COFD, left axis) for different combinations of predictors, ranked by their performance in terms of the COFD: the predictors used for each one model (rows) are marked with grey boxes. The best model with the highest COFD uses all predictors except roughness length (topmost row). (b) Scatter plot of the estimated and observed values by the optimum model. Crosses mark estimates of the full calibration; blue dots mark a cross-validation by leaving out data of the site. The station ‘Zugspitze’ is marked with the station number 10961. and intensities. For the comparison with the COSMO- The comparison of the tracks is shown in Table 2 and Fig. CLM results, core pressure is considered as a measure of 5. In Table 2, characteristics of the 10 strongest cyclones for intensity. The COSMO-CLM cyclone tracks are simply the ERA-Interim period from 1989 to 2007 in terms of constructed from minimum pressure near the ERA-Interim potential damage over Germany (cf. Pinto et al., 2007a; cyclone track, which is sufficient, as the number of tracked Fink et al., 2009) calculated from reanalysis data are cyclones within the RCM domain is limited, and the track compared for reanalysis and COSMO-CLM simulations. can thus be identified unequivocally. Comparison is done Exceptfor Daria (24 January 1990), thecore pressure values only for the segment of the cyclone track within the are in good agreement. Fig. 5 exemplarily shows four COSMO-CLM domain. cyclone tracks following very different paths with different
Table 2. Key features of the tracks of the strongest 10 storms (see text) simulated with CCLM
Storm CCLM ERA-Interim
Date Lat (8N) Lon (8E) Pmin (hPa) Date Lat (8N) Lon (8E) Pmin (hPa)
Daria 25 January 1990 21UTC 56.438N 4.638E 958.02 25 January 1990 18UTC 56.828N 0.428E 949.13 Vivian 27 February 1990 12UTC 61.728N 19.098E 938.86 27 February 1990 12UTC 60.678N 21.148E 941.04 Wiebke 1 March 1990 03UTC 52.468N 11.288E 976.01 1 March 1990 06UTC 52.268N 18.958E 971.8 Verena 14 January 1993 10UTC 58.318N 23.678E 973.68 14 January 1993 06UTC 57.768N 19.538E 973.07 Barbara 24 January 1993 05UTC 59.978N 3.008E 965.43 24 January 1993 00UTC 59.178N 3.748W 966.8 Anatol 4 December 1999 00UTC 57.438N 18.068E 958.15 3 December 1999 18UTC 56.968N 9.678E 956.42 Lothar 27 December 1999 00UTC 51.398N 22.828E 974.75 26 December 1999 12UTC 50.468N 9.378E 976.09 Jeanett 27 October 2002 14UTC 56.328N 7.068E 977.86 27 October 2002 12UTC 56.448N 4.058E 975.32 Kyrill 19 January 2007 02UTC 56.478N 24.018E 962.97 19 January 2007 06UTC 56.008N 28.548E 961.51 Emma 29 February 2008 21UTC 62.728N 1.148W 956.45 29 February 2008 18UTC 62.348N 4.668W 959.97
Shown is the date and time, at which the minimum sea level pressure Pmin occurred, the geographical position and the minimum pressure value. The storms are in chronological order. 10 K. BORN ET AL.
Fig. 5. Storm tracks, storm footprints (maximum wind gust speed during the event) and series of minimum pressure for four of the strongest storm events simulated with the COSMO-CLM (green tracks, colour-shaded gust speed in m s 1), in comparison to ERA-Interim Reanalysis (black tracks). The lower panels show time series of sea level pressure in hPa, x-axis is longitude. The dots mark six-hourly steps, which is the resolution of ERA-Interim, but COSMO-CLM tracks have been drawn hourly. All tracks were limited to the parts that lie entirely inside the COSMO-CLM domain. (a) Daria, 25 January 1990, (b) Verena, 14 January 1993, (c) Lothar, 26 December 1999 and (d) Kyrill, 18 January 2007.
intensities and characteristics (Daria, Barbara, Lothar, 4.3. Comparison of various WGE formulations for Kyrill). Results document that the tracks are generally in single storms very good agreement. However, and particularly for cases when the track includes open systems during life-time, that In this paragraph, results of WGE methods are compared. means a vorticity minimum without closed isobars (like, for Fig. 6 shows footprints of storm ‘Anatol’ (3 December example, Lothar) on the reanalysis grid, the tracks may 1999; cf. Ulbrich et al., 2001). These footprints depict the differ considerably, which does notcome unexpectedly. maximum wind gustfor each model grid pointduring the WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 11
Fig. 6. Patterns of WGE for storm Anatol (a) WGE Brasseur, (b) WGE DWD and (c) WGE TKE. For further details, see text. whole storm episode, thereby providing a wind gust interval of the WGE, marked by the difference of the ‘signature’ of the storm. Comparing the panels Fig. 6a c, quantiles q05 and q95, is typically 10 m s 1, reaching also the WGE Brasseur (Fig. 6a) estimates highest wind speeds values of 20 m s 1 at mountain sites (10961 Zugspitze, with little land sea differences, while the two other 10980 Wendelstein), and sometimes at coastal stations methods provide very similar patterns (Fig. 6b, c). This is (10129 Bremerhaven, 10147 Hamburg). The spread of the the case for the area primarily affected by the cyclone gusts uncertainty range depends on the average turbulent pffiffiffiffiffi (North Sea, Denmark and Baltic Sea) and nearby areas wind speed vturb ¼ v þ 2q , therefore, it is varying in both (e.g. Germany). Over water, differences between WGE time and space. With respect to possible damage estimation methods are smaller. Over land, WGE Brasseur shows less from WGE, the large uncertainty indicates that the con- reduction in gust speed and, thus, estimates higher gusts sideration of probabilistic aspects might be useful. compared with WGE DWD and WGE TKE. An over- estimation of gusts is also apparent in Brasseur (2001) and 4.4. Computation of skill scores for the whole storm seems to be confined to storms, whereas less extreme situations are represented well. sample In Fig. 7, WGE for three exemplary storms and all Next, an overall evaluation of WGE methods is performed available gustobservationsare shown. Also, mean 10 m taking as many historical storms into account as the wind speeds simulated and from observations are depicted, observations allow (up to the end of 2005). For the in order to see if gust over- or under-estimation corresponds calculation of the scores, only maximum wind gusts per to a similar failure in the average wind speed. event were considered, which reduces the effects of temporal As expected from Fig. 6, the WGE Brasseur method phase shifts of a storm event. The three scores aim at three overestimates gusts in high wind speed situations with gusts different aspects of quality: The QSS evaluates the form of larger than 30 m s 1, except at mountain sites, where it fits the gust distribution without any emphasis on the temporal better to the observations. For gust speeds below 30 m s 1, correlation of model data with observations; the RSS this systematic overestimation cannot be seen. On the other quantifies the effect of deviation between model and obser- hand, for storm Lothar (26 December 1999), which had an vations, the correlation CORR only evaluates temporal co- impactfar away from coastalregions in Germany (e.g. incidence. Because QSS and RSS require a reference method Ulbrich et al., 2001), results of WGE Brasseur were in better for comparison, a WGE using a spatially varying, but temp- agreement with the other WGE methods than for storms orary constant gust factor from Fig. 3 is defined as reference moving over the North/Baltic Seas (e.g. Kyrill or Anatol). method. As Fig. 8 shows, the Brasseur-type WGE has a less WGE DWD and the probabilistic estimate with the WGE good performance than the TKE-based WGEs, except at TKE are relativelysimilar and generally, butnotalways, in mountain sites. At some locations, WGE Brasseur is even better concordance with observations. Deficiencies are worse than the constant gust factor. WGE DWD and the mostly related to failures in model prediction, as the probabilistic TKE approach, where only the median value comparison with mean 10 m wind speeds shows: Both has been considered for scores, behave in a very simi- WGE DWD and the probabilistic WGE TKE approach lar way. Overall, the WGE DWD shows in this study fail if the mean 10 m wind is not predicted correctly (e.g. slightly better skill scores than the other approaches (Table station 10980 for storm Lothar). The width of the 90% 3), although the difference to the probabilistic WGE TKE 12 K. BORN ET AL.
Fig. 7. Wind speed of gusts and 10 m winds at all available observational sites for three exemplary storms, Anatol (3 December 1999), Lothar (26 December 1999) and Jeanett (27 October 2002). The standard WGE methods after Brasseur (2001) (WGE Brasseur) and Schulz and Heise (2003) (WGE DWD) are compared with the TKE-based probabilistic estimation (box and whiskers for 5, 25, 50, 75 and 95% quantiles, respectively). The difference between 5 and 95% quantiles mark the range in which 90% of gusts are expected to occur. The latter are slightly shifted for easier comparison. For more details on each station, see Table 1. is due to the same physical base of both approaches very WGE (not shown) and provides more comparable results to small. The good performance of the WGE DWD could be the other methods. expected, as this method was developed for Germany by the DWD. It has been slightly tuned by choosing 30 m instead of 5. Discussion 10 m in the original formulation as reference height for available momentum and TKE in the Prandtl-layer of the Our results indicate that the three different WGE ap- model. Even though the WGE Brasseur method performs, proaches may provide quite diverse results. However, a in general, less well in this comparison, it has to be stated main finding is that the WGE Brasseur approach produces that the potential of fine-tuning has not been performed results, which differ from the other two methods. Further, for this study. The consideration of a changed numerical WGE DWD and WGE TKE deliver very similar gust implementation may counteract the overestimation of this patterns and time series. Such behaviour could be expected, WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 13
Fig. 8. Skill scores for the quality of the statistical distributions of gusts (QSS), the deviation of gust estimates from observations (RSS) and for temporal coincidence (CORR) at climate observation sites in Germany for WGE Brasseur (light grey), WGE TKE (dark grey) and WGE DWD (black). On the last row, an average over all stations per approach is given. For each station, the maximum number of considered storms is limited by availability of observations. M indicates a mountain station (height above 800 m a.s.l.). For more details on each station, see Table 1. as the WGE Brasseur is in general methodically different However, although fine-tuning for WGE parameters and from the others. WGE Brasseur overestimates wind gusts formulation of the discretisation has not been performed in flat terrain, whereas skill scores even suggest a better extensively in this study, results indicate that the quality of performance at mountain sites (cf. also Pinto et al., 2009). WGE may be improved by further calibration. From this 14 K. BORN ET AL.
Table 3. Averaged skill scores for all stations and all events using gusts. Differences in temporal behaviour are reduced by investigated WGE methods considering footprints of storms, that is, the maximum gusts during the storm period, instead of hourly values for DWD TKE Brasseur calculation of skill scores. QSS 0.69 0.63 0.24 One of the main advantages of the WGE TKE is RSS 0.63 0.57 0.96 the consideration of a probabilistic formulation and, CORR 0.65 0.65 0.60 thus, of a measure of uncertainty for each value. For example, the 90% uncertainty intervals range from around See text for details on skill scores and different WGE formulations. 10 m s 1 in average to 25 m s 1 atmountainand some coastal stations, making clear that probabilistic interpreta- point of view, a general quality statement on the methods tion of possible wind-related damages can be important. may be debatable; only the actual realisation (in our case Thus, such an approach, including a probabilistic assess- an implementation in the COSMO-CLM model) can be ment of uncertainty ranges, may be of added value not only rated. for issuing appropriate severe weather warnings, but also Due to their intrinsic characteristics, WGE Brasseur and for application for wind-related damage estimation (e.g. WGE DWD can be applied in every grid cell of an NWP Pinto et al., 2007a, 2010; Della-Marta et al., 2010; Schwierz model and are able to deliver high-resolution estimates of et al., 2010) and wind energy estimates (e.g. Barthelmie gust patterns. Nevertheless, the calibration evaluation is etal., 2008; Pryor and Barthelmie,2010). confined to observational sites; also for the WGE TKE, the probabilistic assessment of uncertainty ranges is based on local observations. The spatial interpolation of WGE TKE 6. Summary and conclusions is in principle possible, but using less sophisticated The present study compares three WGE methods with approaches simple MLRs using fixed topographic respect to their forecast quality using different skill scores characters as predictors it provides not satisfying results. representing the similarity of probability distributions, the Although statistical characteristics of the distribution are standard error and the temporal correlation. Two of the expected to depend very much on local topographic effects WGE methods estimate gusts locally from mean wind related to land cover (in terms of roughness length) or speed and the turbulence state of the atmosphere (WGE exposition, height and land-use in the nearest region of the DWD and WGE TKE), the third one named after Brasseur observation sites (among other factors), dynamic factors (2001) represents a mixing-down of high momentum within like prevailing wind direction leading to advection of TKE the PBL. The proposed WGE TKE permits a probabilistic and, of course, TKE itself seem to be more important for a interpretation using statistical characteristics of gusts at predictive skill of a spatial interpolation model. All these observational sites for an assessment of uncertainty. The factors are potential predictors in a multiple, not necessa- WGE methods are implemented in the regional climate rily linear, regression model, which would have to be model, COSMO-CLM, which has been applied to 158 applied within the atmospheric model. An ‘offline’ version windstorms of the last four decades. The WGE methods of a MLR model, which takes four topographic character- are applied for each time step, calculating the maximum istics into account but which neglects dynamic forcing, is gust during every output interval. WGEs are compared not a satisfying option to spatially interpolate uncertainties with gust observations at 37 observational sites in in terms of the width of local gust factor distributions (see Germany. Section 4.1). A satisfying interpolation technique (similar In terms of all skill scores, the two local WGE methods to Haas and Born, 2011), considering further dynamical show an overall better behaviour. WGE Brasseur shows parameters, requires far more attention than the present hardly a reduction of gust wind speeds over land compared article can provide. Therefore, an interpolation of the with sea, leading to an overestimation between gusts over statistical characteristics of gustiness between observational flatland and moderately hilly regions. The Brasseur method sites is not provided here and is left for future work. has only better skill scores for mountain stations and in As already stated, the WGE TKE method and the WGE situations with weaker winds. The potential of fine-tuning DWD implementation behave very similar in terms of the has not been applied in this study. In fact, extensive skill scores. The time series of observed and simulated wind calibration and theoretical superiority may be competing speeds indicate that gusts cannot be predicted correctly if effects: a theoretically more appropriate method may be the NWP model already underestimates mean wind speeds. worse in practice than any well fitted approach. Relatively small displacements of wind patterns, for For historical reasons, a lot of WGE methods do not example, connected with the cold front passage, result in take TKE into account directly. The results of the present large discrepancies between observations and simulated study document that using TKE as parameter for gust WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 15
estimation is especially valuable for NWP models, which in average and subscale portions of a variable, e. g. 0 supply TKE as prognostic or diagnostic variable. Without ui ¼ u i þ ui, the mean kinetic energy E consists of one extensive tuning, WGE TKE is able to predict gusts at a term caused by average winds and another term caused by comparable quality as the WGE DWD method. For cases wind deviations. Using Einstein’s summation convention when no TKE can be used directly or in a diagnostic way, and the definition of average TKE: estimates of atmospheric static stability may provide better 1 results than constant gust factors. However, physically q :¼ uiui; (A.1) based methods should be preferred. The TKE formulation 2 has the advantage that it allows for a ‘native’ interpretation E can be expressed in terms of the kinetic energy of the of wind gusts as a result of local TKE. Thus, we propose 1 , mean wind speed ðu iÞ¼2 u iu i and q : that the consideration of a probabilistic WGE TKE approach in NWP models may have several advantages E ¼ Eðu iÞþq (A.2) towards other methods, particularly as it allows for an Let ðvmaxÞi pbeffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the components of the wind gust vector estimation of uncertainties. and vmax ¼ ðvmaxÞiðvmaxÞi the wind gust speed, then the The WGE TKE method introduced in this work does not 0 1 0 0 definitions ðvmaxÞi :¼ðvmaxÞi À u i and qmax :¼ 2 ðvmaxÞiðvmaxÞi consider either fine-tuning or spatial interpolation. While lead to the following decomposition of the maximum the fine-tuning may not be of general interest, as its kinetic energy available for gusts: usefulness may be restricted to the fitted region and the pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi particular NWP model characteristics, the spatial interpola- 1 1 2 E ¼ ðv Þ ðv Þ ¼ 2Eðu Þ þ 2q (A.3) tion may be valuable for an improvement of gust estima- max 2 max i max i 2 i max tions in regions with insufficient observations. Because of The maximum gust speeds are expected to occur when the unknown portion of the impact of local topographic mean wind and gust vectors have the same direction. characteristics, this interpolation has to be carried out very Expressing vmax in terms of Emax yields: carefully and will be the objective of future work. pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi vmax ¼ 2Emax ¼ 2Eðu iÞ þ 2qmax (A.4) 7. Acknowledgements In eq. (A.4), qmax may be expressed in terms of the known This research has been funded by the German Association grid-scale TKE and an unknown, subscale stochastic part. of Insurers (‘Gesamtverband der Deutschen Versicherungs- Thus, using v as average wind speed, eq. (A.4) may be wirtschaft’, GDV) in a project dealing with the impacts of rewritten as: climate change for the insurance industry for Germany pffiffiffiffiffi v v 2q e (A.5) (‘Auswirkungen des Klimawandels auf die Schadensitua- max ¼ þ þ v tion in der deutschen Versicherungswirtschaft’). Model with ov being the square root of the difference between the simulations have been performed at the Computing Centre energy of the wind speed deviation v? and the TKE: of the Cologne University (RRZK) and the German Climate Computing Centre (DKRZ). We thank the Eur- 1 2 ev ¼ qmax À q (A.6) opean Centre for Medium Range Weather Forecast 2 (ECMWF, UK) for reanalysis data and the German Equation (A.5) is a key equation for turbulence-driven gust Weather Service (DWD) for providing synoptic station parameterisations, as they all can be expressed using this data. We thank Rabea Haas for helping to prepare Fig. 4 formula. It is an advantageous formulation for most state- and Sven Ulbrich (both Univ. Cologne) for Fig. 5 and of-the art mesoscale models, as TKE is usually a prognostic Table 2. variable of the turbulence parameterisation. Equation (A.4)
is exact, if ov is known, which is variable in time and space. 8. Appendix A: The gustfactor( gv) can then be written as: pffiffiffiffiffi 2q A.1. Basic derivation of turbulence-driven wind gust g ¼ 1 þ þ e (A.7) v v g estimation methods The random parts o and e ¼ e = v are also variable both in We propose the use of the near-surface TKE for analysing v g v space and time. In the WGE DWD, eq. (A.5) is approxi- the relation between average wind speed and wind maxima. mated using: This approach is similar to the theory proposed by Wichers pffiffiffiffiffi Schreur and Geertsema (2008), but it handles the TKE in a 2q þ ev auà differentway. Following Reynolds’ conceptof separation 16 K. BORN ET AL.
with a semi-empirical factor a, based partly on PBL theory type SS is zero for equal quality of both methods; for considerations (Panofsky and Dutton, 1984) and partly values below 0, the evaluated method is worse than the being empirical (see Schulz, 2008; Schulz and Heise, 2003). reference, and for values larger than 0, the tested method is In WGE TKE, the random part is estimated using the gust better than reference with optimum performance at 1. observations. Both WGE DWD and WGE TKE inter- For the RMSE skill score RSS, (o, oref) are rootmean polate v, u* and q to a level of 30 m above surface. squared deviations of WGE and gust observations: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u u XN A.2. Probabilistic approach of WGE t 1 e ¼ ðWGE À v Þ2 (A.11) RSS N i max;i The simplest way to achieve information about wind gust i¼1 distributions is to estimate the width of the WGE distribu- tion using mean wind speed q dependent quantile functions, WGE is the wind gust estimation after one of the three which may be assessed by quantile regression. For that methods and vmax represents gust observations. The idea is purpose, we assume the gust distribution and, thus, the simply that a better WGE should produce less deviation relation between gust factors and mean wind speed to be of between observed and predicted wind gusts. For the exponential power-law type: quantile skill score (QSS), (o, oref) is the sum of distances of points of ranked time series (WGErank, vmax,rank) from b gv ¼ 1 þ expða Á v Þ (A.8) the line of identity in a scatter plot: The assumed type of the fit function does not affect the XN results considerably, as long as curvature, slope and 1 ffiffiffi eQSS ¼ p absððWGEiÞrank Àðvmax;iÞrank (A.12) intercept are used in the fit. Here, very small or large N 2 i¼1 values are discarded due to data availability, as (1) wind gusts are only reported above 12 m s 1; and (2) for some The scaling factor just indicates that in a scatter diagram stations, the largest values are limited to 50 m s 1. The fit of ranked values the length of the shortest path from the of eq. (A.8) can be undertaken via linear regression using: point( WGErank, vmax,rank) to the line of identity is measured. The QSS evaluates the form of distributions: lnðlnðg À 1ÞÞ ¼ lnðaÞþb Á lnð vÞ (A.9) v although temporal correlation may be poor, the ranked Equation (A.9) allows for an estimation of parameters b events can be similar in a scatter plot. and a by linear quantile regression, which gives an assessment of the form of gust distributions at constant References mean wind speed by showing 5, 25, 50, 75 and 95% quantiles (q05, q25, q50, q75 and q95). Agustsson, H. and Olafsson, H. 2004. Mean gust factors over complex terrain. Meteorol. Z. 13, 149 155. Agustsson, H. and Olafsson, H. 2009. Forecasting wind gusts in A.3. Skill scores complex terrain. Meteor. Atmos. Phys. 103, 173 185. Barthelmie, R. J., Murray, F. and Pryor, S. C. 2008. The economic The evaluation of the WGE is undertaken using skill benefit of short-term forecasting for wind energy in the UK scores. The first and most simple score is the temporal electricity market. Energy Policy 36, 1687 1696. correlation CORR of WGE and observations at weather Bo¨ hm, U., Keuler, K., O¨ sterle, H., Ku¨ cken, M. and Hauffe, D. stations for storm episodes. It reflects the temporal 2008. Quality of a climate reconstruction for the CADSES accordance of the two time series without regard to the region. MetZ. Spec. Iss. Regional Clim. Model. COSMO-CLM absolute values. The other two scores are formulated in (CCLM) 17(8), 477 485. analogue to the Brier skill score and are designed to Brasseur, O. 2001. Developmentand applicationof a physical compare a method in focus with a reference method. The approach to estimating wind gusts. Mon. Wea. Rev. 129,5 25. reference method is the WGE with a spatially varying but Brasseur, O., Gallee, H., Boyen, H. and Tricot, C. 2002. Reply. temporarily constant gust factor obtained from observa- Mon. Wea. Rev. 130, 1936 1943. Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, tions (see Fig. 3); the compared methods are either WGE P. and co-authors. 2011. The ERA-Interim reanalysis: Brasseur, the WGE DWD or WGE TKE. The basic form configuration and performance of the data assimilation system. of all Brier-type skill scores is: Quart. J. R. Meteor. Soc. 137, 553 597. DOI: 10.1002/qj.828. e Della-Marta, P. M., Liniger, M. A., Appenzeller C., Bresch D. N., SSðeÞ¼1 À (A.10) Ko¨ llner-Heck P. and Muccione V. 2010. Improved estimates of eref the European winter wind storm climate and the risk of with different types of error estimates (o, oref) for WGE reinsurance loss using climate model data. J. Appl. Meteor. methods and the reference method, respectively. A Brier- Clim. 49, 2092 2120. WIND GUST ESTIMATION FOR MID-EUROPEAN WINTER STORMS 17
Della-Marta, P. M., Mathis, H., Frei, C., Liniger, M. A., Kleinn, J. modelling of storm events over West Germany. Meteorol. Z. 18, and Appenzeller, C. 2009. The return period of windstorms over 495 506. Europe. Int. J. Climatol. 29, 437 459. Pinto, J. G., Neuhaus, C. P., Leckebusch, G. C., Reyers, M. and De Rooy, W. C. and Kok, K. 2004. A combined physical statistical Kerschgens, M. 2010. Estimation of wind storm impacts over approach for the downscaling of model wind speed. Wea. Forec. West Germany under future climate conditions using a statis- 19, 485 495. tical-dynamical downscaling approach. Tellus A 62, 188 201. Durst, C. D. 1960. Wind speeds over short periods of time. Pinto, J. G., Spangehl, T., Ulbrich, U. and Speth, P. 2005. Meteorol. Mag. 89, 181 186. Sensitivities of a cyclone detection and tracking algorithm: Fink, A. H., Bru¨ cher, T., Ermert, E., Kru¨ ger, A. and Pinto, J. G. individual tracks and climatology. Meteorol. Z. 14, 823 838. 2009. The European Storm Kyrill in January 2007: synoptic Pinto, J. G., Ulbrich, U., Leckebusch, G. C., Spangehl, T., Reyers, evolution and considerations with respect to climate change. M. and Zacharias, S. 2007b. Changes in storm track and cyclone Nat. Hazards Earth Syst. Sci. 9, 405 423. activity in three SRES ensemble experiments with the EC- Friederichs, P., Goeber, M., Bentzien, S., Lenz, A. and Krampitz, HAM5/MPI-OM1 GCM. Clim. Dyn. 29, 195 210. R. 2009. A probabilistic analysis of wind gusts using extreme Pryor, S. C. and Barthelmie, R. J. 2010. Climate change impacts value statistics. Meteorol. Z. 18, 615 629. on wind energy: a review. Renewable and Sustainable Energy Goyette, S., Brasseur, O. and Beniston, M. 2003. Application of a Reviews 14, 430 437. new wind gust parameterisation: multiple scale studies per- Ritter, B. and Geleyn, J.-F. 1992. A comprehensive radiation formed with the Canadian regional climate model. J. Geophys. scheme of numerical weather prediction with potential applica- Res. 108(D13), 4374. DOI: 10.1029/2002JD002646. tion to climate simulations. Mon. Wea. Rev. 120, 303 325. Haas, R. and Born, K. 2011. Probabilistic downscaling of Rockel, B., Will, A. and Hense, A. (eds.) (2008). Special issue: precipitation data in a subtropical mountain area: a two-step regional climate modelling with COSMO-CLM (CCLM), approach. Nonlin. Process. Geophys. 18, 223 234. Meteorol. Z. 17, 347 348. Jaeger, E. B., Anders, I., Lu¨ thi, D., Rockel, B., Scha¨ r, C. and co- Schulz, J.-P. 2008. Revision of the turbulent gust diagnostics in the authors. 2008. Analysis of ERA40-driven CLM simulations for COSMO model. COSMO Newslett. 8,17 22. Online at: www. Europe, Meteorol. Z. 17, 349 367. cosmo-model.org Klawa, M. and Ulbrich, U. 2003. A model for the estimation of Schulz, J.-P. and Heise, E. 2003. A new scheme for diagnosing storm losses and the identification of severe winter storms in near-surface convective gusts. COSMO Newslett. 3, 221 225. Germany. Nat. Hazards Earth Syst. Sci. 3, 725 732. Online at: www.cosmo-model.org Lautenschlager, M., Keuler, K., Wunram, C., Keup-Thiel, E., Schwierz, C., Ko¨ llner-Heck, P., Zenklusen Mutter, E., Bresch, Schubert, M., Will, A., Rockel, B. and Boehm, U. (2009). D.N., Vidale, P.-L., Wild, M. and Scha¨ r, C. 2010. Modelling Climate simulation with CLM, scenario A1B run no.1, data European winter wind storm losses in current and future climate. stream 3: European region MPI-M/MaD. World Data Center Clim. Change 101, 485 514. DOI: 10.1007/s10584-009-9712-1. Clim. DOI: 10.1594/WDCC/CLM_A1B_1_D3. Simmonds, I., Murray, R. J. and Leighton, R. M. 1999. A Mellor, G. and Yamada, T. 1982. Developmentof a turbulence refinement of cyclone tracking methods with data from FROST. closure model for geophysical fluid problems. Rev. Geophys. Aust. Met. Mag. Special Edition,35 49. Space Phys. 20, 851 875. Steppeler, J., Dom, G., Scha¨ ttler, U., Bitzer, H. W., Gassmann, A. Murray, R. J. and Simmonds, I. 1991. A numerical scheme for and co-authors. 2003. Meso-gamma scale forecasts using the tracking cyclone centres from digital data. Part I: development nonhydrostatic model LM. Meteorol. Atmos. Phys. 82,75 96. and operation of the scheme. Aust. Meteorol. Mag. 39, 155 166. Ulbrich, U., Fink, A. H., Klawa, M. and Pinto, J. G. 2001. Three Nissen, K. M., Leckebusch, G. C., Pinto, J. G., Renggli, D., extreme storms over Europe in December 1999. Weather 56, Ulbrich, S. and Ulbrich, U. 2010. Cyclones causing wind storms 70 80. in the Mediterranean: characteristics, trends and links to large- Uppala, S. M., Kallberg, P., Hernandez, A., Saarinen, S., scale patterns. Nat. Hazards Earth Syst. Sci. 10, 1379 1391. Fiorino, M. and co-authors. 2005. The ERA-40 reanalysis. Panofsky, H. A. and Dutton, J. A. 1984. Atmospheric Turbulence: Quart. J. R. Meteor. Soc. 131, 2961 3012. Models and Methods for Engineering Applications. John Wiley & Verkaik, J. W. 2000. Evaluation of two gustiness models for Sons, New York, 397 pp. exposure correction values. J. Appl. Meteor. 39, 1613 1626. Pinto, J. G., Fro¨ hlich, E. L., Leckebusch, G. C. and Ulbrich, U. Wichers Schreur, B. and Geertsema, G. 2008. Theory for a TKE 2007a. Changes in storm loss potentials over Europe under based parameterization of wind gusts. HIRLAM Newslett. 54, modified climate conditions in an ensemble of simulations 177 188. Online at: http://hirlam.org/index.php?option com_ of ECHAM5/MPI-OM1. Nat. Hazards Earth Syst. Sci. 7, docman&Itemid